projects / fio.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
[PATCH] Document how to setup/use huge pages
[fio.git] / fio.h
... / ...
CommitLineData
1#ifndef FIO_H
2#define FIO_H
3
4#include <sched.h>
5#include <limits.h>
6#include <pthread.h>
7#include <sys/time.h>
8#include <sys/resource.h>
9#include <errno.h>
10#include <stdlib.h>
11#include <stdio.h>
12#include <unistd.h>
13#include <string.h>
14
15#include "list.h"
16#include "md5.h"
17#include "crc32.h"
18#include "arch.h"
19#include "os.h"
20
21enum fio_ddir {
22 DDIR_READ = 0,
23 DDIR_WRITE,
24 DDIR_SYNC,
25};
26
27struct io_stat {
28 unsigned long val;
29 unsigned long val_sq;
30 unsigned long max_val;
31 unsigned long min_val;
32 unsigned long samples;
33};
34
35struct io_sample {
36 unsigned long time;
37 unsigned long val;
38 enum fio_ddir ddir;
39};
40
41struct io_log {
42 unsigned long nr_samples;
43 unsigned long max_samples;
44 struct io_sample *log;
45};
46
47struct io_piece {
48 struct list_head list;
49 struct fio_file *file;
50 unsigned long long offset;
51 unsigned int len;
52 enum fio_ddir ddir;
53};
54
55/*
56 * The io unit
57 */
58struct io_u {
59 union {
60#ifdef FIO_HAVE_LIBAIO
61 struct iocb iocb;
62#endif
63#ifdef FIO_HAVE_POSIXAIO
64 struct aiocb aiocb;
65#endif
66#ifdef FIO_HAVE_SGIO
67 struct sg_io_hdr hdr;
68#endif
69 };
70 struct timeval start_time;
71 struct timeval issue_time;
72
73 void *buf;
74 unsigned int buflen;
75 unsigned long long offset;
76
77 unsigned int resid;
78 unsigned int error;
79
80 enum fio_ddir ddir;
81
82 /*
83 * io engine private data
84 */
85 union {
86 unsigned int index;
87 unsigned int seen;
88 };
89
90 struct fio_file *file;
91
92 struct list_head list;
93};
94
95#define FIO_HDR_MAGIC 0xf00baaef
96
97enum {
98 VERIFY_NONE = 0,
99 VERIFY_MD5,
100 VERIFY_CRC32,
101};
102
103struct verify_header {
104 unsigned int fio_magic;
105 unsigned int len;
106 unsigned int verify_type;
107 union {
108 char md5_digest[MD5_HASH_WORDS * 4];
109 unsigned long crc32;
110 };
111};
112
113struct group_run_stats {
114 unsigned long long max_run[2], min_run[2];
115 unsigned long long max_bw[2], min_bw[2];
116 unsigned long long io_kb[2];
117 unsigned long long agg[2];
118};
119
120/*
121 * What type of allocation to use for io buffers
122 */
123enum fio_memtype {
124 MEM_MALLOC = 0, /* ordinary malloc */
125 MEM_SHM, /* use shared memory segments */
126 MEM_SHMHUGE, /* use shared memory segments with huge pages */
127 MEM_MMAP, /* use anonynomous mmap */
128 MEM_MMAPHUGE, /* memory mapped huge file */
129};
130
131/*
132 * The type of object we are working on
133 */
134enum fio_filetype {
135 FIO_TYPE_FILE = 1,
136 FIO_TYPE_BD,
137 FIO_TYPE_CHAR,
138};
139
140enum fio_ioengine_flags {
141 FIO_SYNCIO = 1 << 0,
142 FIO_CPUIO = 1 << 1,
143 FIO_MMAPIO = 1 << 2,
144 FIO_RAWIO = 1 << 3,
145};
146
147struct fio_file {
148 /*
149 * A file may not be a file descriptor, let the io engine decide
150 */
151 union {
152 unsigned long file_data;
153 int fd;
154 };
155 char *file_name;
156 void *mmap;
157 unsigned long long file_size;
158 unsigned long long real_file_size;
159 unsigned long long file_offset;
160 unsigned long long last_pos;
161 unsigned long long last_completed_pos;
162
163 unsigned long *file_map;
164 unsigned int num_maps;
165
166 unsigned int unlink;
167};
168
169/*
170 * This describes a single thread/process executing a fio job.
171 */
172struct thread_data {
173 char *name;
174 char *directory;
175 char *filename;
176 char verror[80];
177 pthread_t thread;
178 int thread_number;
179 int groupid;
180 enum fio_filetype filetype;
181 struct fio_file *files;
182 unsigned int nr_files;
183 unsigned int nr_uniq_files;
184 unsigned int next_file;
185 int error;
186 pid_t pid;
187 char *orig_buffer;
188 size_t orig_buffer_size;
189 volatile int terminate;
190 volatile int runstate;
191 enum fio_ddir ddir;
192 unsigned int iomix;
193 unsigned int ioprio;
194 unsigned int last_was_sync;
195
196 unsigned int sequential;
197 unsigned int odirect;
198 unsigned int invalidate_cache;
199 unsigned int create_serialize;
200 unsigned int create_fsync;
201 unsigned int end_fsync;
202 unsigned int sync_io;
203 unsigned int verify;
204 unsigned int use_thread;
205 unsigned int unlink;
206 unsigned int do_disk_util;
207 unsigned int override_sync;
208 unsigned int rand_repeatable;
209 unsigned int write_lat_log;
210 unsigned int write_bw_log;
211 unsigned int norandommap;
212 unsigned int bs_unaligned;
213
214 unsigned int bs[2];
215 unsigned int min_bs[2];
216 unsigned int max_bs[2];
217 unsigned int rw_min_bs;
218 unsigned int thinktime;
219 unsigned int fsync_blocks;
220 unsigned int start_delay;
221 unsigned long timeout;
222 unsigned int overwrite;
223 unsigned int bw_avg_time;
224 unsigned int loops;
225 unsigned long long zone_size;
226 unsigned long long zone_skip;
227 enum fio_memtype mem_type;
228 char *hugefile;
229 int hugefd;
230 unsigned int stonewall;
231 unsigned int numjobs;
232 unsigned int iodepth;
233 os_cpu_mask_t cpumask;
234 unsigned int iolog;
235 unsigned int read_iolog;
236 unsigned int rwmixcycle;
237 unsigned int rwmixread;
238 unsigned int rwmixwrite;
239 unsigned int nice;
240
241 char *read_iolog_file;
242 char *write_iolog_file;
243 void *iolog_buf;
244 FILE *iolog_f;
245
246 char *sysfs_root;
247 char *ioscheduler;
248
249 os_random_state_t bsrange_state;
250 os_random_state_t verify_state;
251
252 int shm_id;
253
254 /*
255 * IO engine hooks, contains everything needed to submit an io_u
256 * to any of the available IO engines.
257 */
258 struct ioengine_ops *io_ops;
259
260 /*
261 * Current IO depth and list of free and busy io_u's.
262 */
263 unsigned int cur_depth;
264 struct list_head io_u_freelist;
265 struct list_head io_u_busylist;
266
267 /*
268 * Rate state
269 */
270 unsigned int rate;
271 unsigned int ratemin;
272 unsigned int ratecycle;
273 unsigned long rate_usec_cycle;
274 long rate_pending_usleep;
275 unsigned long rate_bytes;
276 struct timeval lastrate;
277
278 unsigned long runtime[2]; /* msec */
279 unsigned long long io_size;
280 unsigned long long total_file_size;
281 unsigned long long start_offset;
282 unsigned long long total_io_size;
283
284 unsigned long long io_blocks[2];
285 unsigned long long io_bytes[2];
286 unsigned long long zone_bytes;
287 unsigned long long this_io_bytes[2];
288 volatile int mutex;
289
290 /*
291 * State for random io, a bitmap of blocks done vs not done
292 */
293 os_random_state_t random_state;
294
295 /*
296 * CPU "io" cycle burner
297 */
298 unsigned int cpuload;
299 unsigned int cpucycle;
300
301 /*
302 * bandwidth and latency stats
303 */
304 struct io_stat clat_stat[2]; /* completion latency */
305 struct io_stat slat_stat[2]; /* submission latency */
306 struct io_stat bw_stat[2]; /* bandwidth stats */
307
308 unsigned long long stat_io_bytes[2];
309 struct timeval stat_sample_time[2];
310
311 struct io_log *slat_log;
312 struct io_log *clat_log;
313 struct io_log *bw_log;
314
315 struct timeval start; /* start of this loop */
316 struct timeval epoch; /* time job was started */
317 struct timeval end_time;/* time job ended */
318
319 /*
320 * fio system usage accounting
321 */
322 struct rusage ru_start;
323 struct rusage ru_end;
324 unsigned long usr_time;
325 unsigned long sys_time;
326 unsigned long ctx;
327
328 /*
329 * read/write mixed workload state
330 */
331 os_random_state_t rwmix_state;
332 struct timeval rwmix_switch;
333 enum fio_ddir rwmix_ddir;
334
335 /*
336 * Pre-run and post-run shell
337 */
338 char *exec_prerun;
339 char *exec_postrun;
340
341 /*
342 * IO historic logs
343 */
344 struct list_head io_hist_list;
345 struct list_head io_log_list;
346};
347
348#define __td_verror(td, err, msg) \
349 do { \
350 int e = (err); \
351 (td)->error = e; \
352 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, error=%s", __FILE__, __LINE__, (msg)); \
353 } while (0)
354
355
356#define td_verror(td, err) __td_verror((td), (err), strerror((err)))
357#define td_vmsg(td, err, msg) __td_verror((td), (err), (msg))
358
359extern int exitall_on_terminate;
360extern int thread_number;
361extern int shm_id;
362extern int groupid;
363extern int terse_output;
364extern FILE *f_out;
365extern FILE *f_err;
366extern int temp_stall_ts;
367extern unsigned long long mlock_size;
368
369extern struct thread_data *threads;
370
371#define td_read(td) ((td)->ddir == DDIR_READ)
372#define td_write(td) ((td)->ddir == DDIR_WRITE)
373#define td_rw(td) ((td)->iomix != 0)
374
375#define BLOCKS_PER_MAP (8 * sizeof(long))
376#define TO_MAP_BLOCK(td, f, b) ((b) - ((f)->file_offset / (td)->rw_min_bs))
377#define RAND_MAP_IDX(td, f, b) (TO_MAP_BLOCK(td, f, b) / BLOCKS_PER_MAP)
378#define RAND_MAP_BIT(td, f, b) (TO_MAP_BLOCK(td, f, b) & (BLOCKS_PER_MAP - 1))
379
380#define MAX_JOBS (1024)
381
382static inline int should_fsync(struct thread_data *td)
383{
384 if (td->last_was_sync)
385 return 0;
386 if (td->odirect)
387 return 0;
388 if (td_write(td) || td_rw(td) || td->override_sync)
389 return 1;
390
391 return 0;
392}
393
394struct disk_util_stat {
395 unsigned ios[2];
396 unsigned merges[2];
397 unsigned long long sectors[2];
398 unsigned ticks[2];
399 unsigned io_ticks;
400 unsigned time_in_queue;
401};
402
403struct disk_util {
404 struct list_head list;
405
406 char *name;
407 char path[256];
408 dev_t dev;
409
410 struct disk_util_stat dus;
411 struct disk_util_stat last_dus;
412
413 unsigned long msec;
414 struct timeval time;
415};
416
417struct io_completion_data {
418 int nr; /* input */
419
420 int error; /* output */
421 unsigned long bytes_done[2]; /* output */
422 struct timeval time; /* output */
423};
424
425#define DISK_UTIL_MSEC (250)
426
427#ifndef min
428#define min(a, b) ((a) < (b) ? (a) : (b))
429#endif
430#ifndef max
431#define max(a, b) ((a) > (b) ? (a) : (b))
432#endif
433
434/*
435 * Log exports
436 */
437extern int read_iolog_get(struct thread_data *, struct io_u *);
438extern void write_iolog_put(struct thread_data *, struct io_u *);
439extern int init_iolog(struct thread_data *td);
440extern void log_io_piece(struct thread_data *, struct io_u *);
441extern void prune_io_piece_log(struct thread_data *);
442extern void write_iolog_close(struct thread_data *);
443
444/*
445 * Logging
446 */
447extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long);
448extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long);
449extern void add_bw_sample(struct thread_data *, enum fio_ddir, struct timeval *);
450extern void show_run_stats(void);
451extern void init_disk_util(struct thread_data *);
452extern void update_rusage_stat(struct thread_data *);
453extern void update_io_ticks(void);
454extern void disk_util_timer_arm(void);
455extern void setup_log(struct io_log **);
456extern void finish_log(struct thread_data *, struct io_log *, const char *);
457extern int setup_rate(struct thread_data *);
458
459/*
460 * Time functions
461 */
462extern unsigned long utime_since(struct timeval *, struct timeval *);
463extern unsigned long utime_since_now(struct timeval *);
464extern unsigned long mtime_since(struct timeval *, struct timeval *);
465extern unsigned long mtime_since_now(struct timeval *);
466extern unsigned long time_since_now(struct timeval *);
467extern unsigned long mtime_since_genesis(void);
468extern void __usec_sleep(unsigned int);
469extern void usec_sleep(struct thread_data *, unsigned long);
470extern void rate_throttle(struct thread_data *, unsigned long, unsigned int, int);
471extern void fill_start_time(struct timeval *);
472extern void fio_gettime(struct timeval *, void *);
473
474/*
475 * Init functions
476 */
477extern int parse_options(int, char **);
478extern int init_random_state(struct thread_data *);
479
480/*
481 * File setup/shutdown
482 */
483extern void close_files(struct thread_data *);
484extern int setup_files(struct thread_data *);
485extern int open_files(struct thread_data *);
486extern int file_invalidate_cache(struct thread_data *, struct fio_file *);
487
488/*
489 * ETA/status stuff
490 */
491extern void print_thread_status(void);
492extern void print_status_init(int);
493
494/*
495 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
496 * will never back again. It may cycle between running/verififying/fsyncing.
497 * Once the thread reaches TD_EXITED, it is just waiting for the core to
498 * reap it.
499 */
500enum {
501 TD_NOT_CREATED = 0,
502 TD_CREATED,
503 TD_INITIALIZED,
504 TD_RUNNING,
505 TD_VERIFYING,
506 TD_FSYNCING,
507 TD_EXITED,
508 TD_REAPED,
509};
510
511/*
512 * Verify helpers
513 */
514extern void populate_verify_io_u(struct thread_data *, struct io_u *);
515extern int get_next_verify(struct thread_data *td, struct io_u *);
516extern int do_io_u_verify(struct thread_data *, struct io_u **);
517
518/*
519 * Memory helpers
520 */
521extern int fio_pin_memory(void);
522extern void fio_unpin_memory(void);
523extern int allocate_io_mem(struct thread_data *);
524extern void free_io_mem(struct thread_data *);
525
526/*
527 * io unit handling
528 */
529#define queue_full(td) list_empty(&(td)->io_u_freelist)
530extern struct io_u *__get_io_u(struct thread_data *);
531extern struct io_u *get_io_u(struct thread_data *, struct fio_file *);
532extern void put_io_u(struct thread_data *, struct io_u *);
533extern void ios_completed(struct thread_data *, struct io_completion_data *);
534extern void io_completed(struct thread_data *, struct io_u *, struct io_completion_data *);
535
536/*
537 * io engine entry points
538 */
539extern int td_io_init(struct thread_data *);
540extern int td_io_prep(struct thread_data *, struct io_u *);
541extern int td_io_queue(struct thread_data *, struct io_u *);
542extern int td_io_sync(struct thread_data *, struct fio_file *);
543extern int td_io_getevents(struct thread_data *, int, int, struct timespec *);
544
545/*
546 * This is a pretty crappy semaphore implementation, but with the use that fio
547 * has (just signalling start/go conditions), it doesn't have to be better.
548 * Naturally this would not work for any type of contended semaphore or
549 * for real locking.
550 */
551static inline void fio_sem_init(volatile int *sem, int val)
552{
553 *sem = val;
554}
555
556static inline void fio_sem_down(volatile int *sem)
557{
558 while (*sem == 0)
559 usleep(10000);
560
561 (*sem)--;
562}
563
564static inline void fio_sem_up(volatile int *sem)
565{
566 (*sem)++;
567}
568
569/*
570 * If logging output to a file, stderr should go to both stderr and f_err
571 */
572#define log_err(args...) do { \
573 fprintf(f_err, ##args); \
574 if (f_err != stderr) \
575 fprintf(stderr, ##args); \
576 } while (0)
577
578struct ioengine_ops {
579 struct list_head list;
580 char name[16];
581 int version;
582 int flags;
583 int (*setup)(struct thread_data *);
584 int (*init)(struct thread_data *);
585 int (*prep)(struct thread_data *, struct io_u *);
586 int (*queue)(struct thread_data *, struct io_u *);
587 int (*getevents)(struct thread_data *, int, int, struct timespec *);
588 struct io_u *(*event)(struct thread_data *, int);
589 int (*cancel)(struct thread_data *, struct io_u *);
590 void (*cleanup)(struct thread_data *);
591 void *data;
592 void *dlhandle;
593};
594
595#define FIO_IOOPS_VERSION 3
596
597extern struct ioengine_ops *load_ioengine(struct thread_data *, const char *);
598extern int register_ioengine(struct ioengine_ops *);
599extern void unregister_ioengine(struct ioengine_ops *);
600extern void close_ioengine(struct thread_data *);
601
602/*
603 * Mark unused variables passed to ops functions as unused, to silence gcc
604 */
605#define fio_unused __attribute((__unused__))
606#define fio_init __attribute__((constructor))
607#define fio_exit __attribute__((destructor))
608
609#define for_each_td(td, i) \
610 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
611#define for_each_file(td, f, i) \
612 for ((i) = 0, (f) = &(td)->files[0]; (i) < (int) (td)->nr_files; (i)++, (f)++)
613
614#endif
fio - Flexible IO Tester